Multimedia information recommendation method, device and equipment, and computer storage medium

Abstract

The application discloses a multimedia information recommendation method, device and equipment and a computer storage medium, relates to the technical field of information recommendation, and is used for improving the multimedia information recall accuracy and the utilization rate of the precision queue, thereby improving the overall accuracy of the recommendation system. In the method, based on a target recommendation request of a target account, at least one historical tail set of the target account is acquired; based on the at least one historical tail set and a set filtering strategy, an information filtering set containing at least one multimedia information is constructed; for the target recommendation request, a recall set containing a plurality of multimedia information is acquired, and the multimedia information included in the information filtering set is screened out from the recall set; based on the screened recall set, a candidate recommendation set corresponding to the target recommendation request is obtained, and target multimedia information recommended for the target account is determined from the candidate recommendation set.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly to the field of information recommendation technology, providing a multimedia information recommendation method, apparatus, device, and computer storage medium. Background Technology

[0002] Recommendation systems generally consist of three stages: recall, coarse ranking, and fine ranking. When recommending multimedia information, each recommendation request involves independent recall and coarse ranking phases. Recall refers to retrieving multimedia information that the user might be interested in from the entire multimedia information database using a multi-path recall approach from different perspectives. Coarse ranking involves efficiently and lightweightly sorting the recalled multimedia information at a coarse granular level, quickly filtering out a small number of relatively high-quality multimedia messages as input for the fine ranking stage. The fine ranking stage then performs a finer-grained ranking, ultimately selecting the optimal multimedia information from various advertisements and recommending it to the user.

[0003] Therefore, for each recommendation request, before entering the fine ranking stage, it needs to go through the recall and coarse ranking stages. Thus, the recall and coarse ranking stages directly determine which multimedia information can enter the fine ranking stage, thereby determining to a certain extent the multimedia information to be recommended. Therefore, the accuracy of the recall and coarse ranking stages determines the overall accuracy of the recommendation system to a certain extent. Summary of the Invention

[0004] This application provides a multimedia information recommendation method, apparatus, device, and computer storage medium to improve the accuracy of multimedia information retrieval and the utilization rate of the fine-ranking queue, thereby improving the overall accuracy of the recommendation system.

[0005] On the one hand, a multimedia information recommendation method is provided, the method comprising:

[0006] Based on the target recommendation request of the target account, at least one historical tail set of the target account is obtained; wherein, each historical tail set includes the last N multimedia information items in the candidate recommendation set corresponding to a historical recommendation request, sorted from high to low according to the recommendation degree, where N is a positive integer;

[0007] Based on the at least one historical tail set and the set filtering strategy, an information filtering set containing at least one multimedia information is constructed.

[0008] For the target recommendation request, a recall set containing multiple multimedia information is obtained, and the multimedia information included in the information filtering set is filtered out from the recall set;

[0009] Based on the filtered recall set, a candidate recommendation set corresponding to the target recommendation request is obtained, and

[0010] From the candidate recommendation set, target multimedia information recommended for the target account is determined.

[0011] On the one hand, a multimedia information recommendation device is provided, the device comprising:

[0012] The acquisition unit is used to acquire at least one historical tail set of the target account based on the target recommendation request of the target account; wherein each historical tail set includes the last N multimedia information items in the candidate recommendation set corresponding to the historical recommendation request, sorted from high to low according to the recommendation degree, where N is a positive integer;

[0013] A construction unit is used to construct an information filter set containing at least one multimedia information based on the at least one historical tail set and the set filtering strategy.

[0014] The filtering unit is used to obtain a recall set containing multiple multimedia information for the target recommendation request, and filter out the multimedia information included in the information filtering set from the recall set;

[0015] The recommendation unit is used to obtain a candidate recommendation set corresponding to the target recommendation request based on the filtered recall set, and to determine the target multimedia information recommended for the target account from the candidate recommendation set.

[0016] Optionally, the construction unit is specifically used for:

[0017] Based on a set duration threshold, a set of historical tails that is in a valid state is determined from the at least one set of historical tails; wherein, when the time difference between the generation time of the historical recommendation request corresponding to a set of historical tails and the current time is not greater than the duration threshold, the set of historical tails is in a valid state.

[0018] Based on the recommendation attribute information of the target account or the target recommendation request, obtain the values ​​of each filtering parameter in the filtering strategy that corresponds to the recommendation attribute information;

[0019] Based on the values ​​of the obtained filtering parameters, at least one multimedia information is selected from the set of historical tails that are in a valid state to form the information filtering set.

[0020] Optionally, the construction unit is specifically used for:

[0021] Based on the recommendation dimension information to which the target recommendation request belongs, the values ​​of each filtering parameter in the filtering strategy corresponding to the recommendation dimension information are obtained; wherein, the recommendation dimension information is used to characterize the source information of the information display position corresponding to the target recommendation request;

[0022] Based on the first recommendation frequency of the target account, obtain the values ​​of each filtering parameter in the filtering strategy corresponding to the first recommendation frequency;

[0023] Based on the second recommendation frequency of the target account in the recommendation dimension to which the target recommendation request belongs, obtain the values ​​of each filtering parameter in the filtering strategy corresponding to the second recommendation frequency.

[0024] Optionally, each filtering parameter includes the number Z of historical recommendation requests corresponding to the historical tail set, and the number M of multimedia information selected from each historical tail set, where M is a positive integer and M≤N; then the construction unit is specifically used for:

[0025] From the set of historical tails that are in a valid state, select the last Z sets of historical recommendation requests that are sorted from largest to smallest by time difference; wherein, the time difference is the time difference between the generation time of the historical recommendation request corresponding to each set of historical tails and the current time.

[0026] From the last Z sets of historical tails, select the last M multimedia information items sorted from high to low according to recommendation level;

[0027] The information filtering set is constructed based on M multimedia information items obtained from each of the Z historical tail sets.

[0028] Optionally, the acquisition unit is specifically used for:

[0029] The target recommendation request is identified by protocol recognition to determine the target site that sent the target recommendation request;

[0030] Based on the parsing method corresponding to the target site, the target recommendation request is parsed to obtain the source information of the information display position corresponding to the target recommendation request;

[0031] Based on the source information of the information display position, the recommendation dimension to which the target recommendation request belongs is determined, and at least one historical tail set is obtained from the historical tail set corresponding to the recommendation dimension.

[0032] Optionally, the recommendation unit is specifically used for:

[0033] Using a trained interaction rate acquisition model, the estimated interaction rate of each multimedia information is obtained based on the resource attribute information of each multimedia information in the candidate recommendation set.

[0034] Based on the obtained estimated interaction rates, the estimated amount of electronic resources that can be obtained after each multimedia message is displayed a set number of times in the information display position is determined.

[0035] The multimedia information is sorted according to the obtained estimated electronic resource quantities from high to low.

[0036] Based on the ranking results, the target multimedia information is determined from the candidate recommendation set.

[0037] Optionally, the apparatus further includes a set update unit for:

[0038] Based on the ranking results, the last N multimedia information items in the candidate recommendation set are selected to form the target tail set corresponding to the target recommendation request;

[0039] The target tail set is added as a historical tail set to the historical tail set library of the target account.

[0040] Optionally, the set update unit is further configured to:

[0041] The relevant information of the target tail set is added to the historical tail set database of the target account; wherein, the relevant information includes one or more of the following:

[0042] The account identifier of the target account;

[0043] The generation time of the target recommendation request;

[0044] The information display position identifier corresponding to the target recommendation request;

[0045] The information display scenario identifier corresponding to the target recommendation request;

[0046] The target site identifier corresponding to the target recommendation request.

[0047] On one hand, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above methods.

[0048] On the one hand, a computer storage medium is provided that stores computer program instructions thereon, which, when executed by a processor, implement the steps of any of the above methods.

[0049] On one hand, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of any of the methods described above.

[0050] In this embodiment, a historical tail set of the target account is collected. Each historical tail set includes tail multimedia information from the candidate recommendation set corresponding to historical recommendation requests. When recommending multimedia information to the target account, an information filtering set is constructed based on at least one historical tail set and a set filtering strategy. After recall, multimedia information included in the information filtering set is filtered out from the recall set, and recommendations are made based on the filtered recall set. Since the tail multimedia information in the candidate recommendation set usually has no chance of being recommended to the user, filtering the tail multimedia information prevents it from occupying candidate recommendation set slots for a long time, giving other multimedia information a chance to enter the candidate recommendation set, thus improving the utilization rate of the candidate recommendation set. Furthermore, the generation of tail multimedia information may be due to an overestimation of its recommendation degree during the recall or coarse-ranking stage. Filtering these multimedia information essentially corrects the predictions made during the recall or coarse-ranking stage, thereby improving the accuracy of multimedia information recall and ultimately improving the overall accuracy of the recommendation system. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 Application scenario diagrams provided for embodiments of this application;

[0053] Figure 2 A schematic diagram of the system architecture of the recommendation server provided in the embodiments of this application;

[0054] Figure 3 A flowchart illustrating the multimedia information recommendation method provided in this application embodiment;

[0055] Figure 4 A schematic diagram of the information display positions provided in the embodiments of this application;

[0056] Figure 5 A schematic diagram of the storage structure of the tail set provided in the embodiments of this application;

[0057] Figure 6 A schematic diagram of the process for obtaining the construction information filtering set provided in an embodiment of this application;

[0058] Figure 7 A schematic diagram illustrating the obtained information filtering set provided in an embodiment of this application;

[0059] Figure 8 A flowchart illustrating the fine-sorting process provided in this application embodiment;

[0060] Figure 9 A schematic diagram illustrating the process of recommending advertisements provided in this application embodiment;

[0061] Figure 10 A schematic diagram of the structure of the multimedia information recommendation device provided in the embodiments of this application;

[0062] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0064] To facilitate understanding of the technical solutions provided in the embodiments of this application, some key terms used in the embodiments of this application will be explained below:

[0065] Multimedia information refers to information that can be displayed through in-app pages or web page display slots, and users can interact with it. Examples include advertisements or web page links. Taking advertisements as an example, web pages or in-app pages have designated ad display slots. Ads can be displayed through these slots, and users can click on the ad to enter the corresponding ad page and perform actions such as downloading the application on the ad page or purchasing the product recommended in the ad.

[0066] Historical Tail Set: The tail refers to the last few items in the ranking. For a recommendation request, multiple multimedia information items will be filtered out. These multimedia information items form a candidate recommendation set. Then, according to a certain ranking strategy, these multimedia information items are ranked. Only the multimedia information items at the top of the ranking will be recommended to the user, while the multimedia information items at the bottom of the ranking will not be recommended to the user. These multimedia information items constitute a historical tail set corresponding to a recommendation set. For example, when the candidate recommendation set includes 100 multimedia information items, if the last 10 items are set as the tail for each recommendation, then the last 10 items out of the 100 items can form a historical tail set.

[0067] Recall set: The recall set refers to the set of multimedia information retrieved from a multimedia information database using recall methods.

[0068] Candidate Recommendation Set: The candidate recommendation set refers to the set of a small number of high-quality multimedia information that have entered the fine-ranking stage.

[0069] Recommendation attribute information: refers to attribute information related to multimedia information recommendation, which can be obtained based on the user's historical recommendation information and the current recommendation scenario information. For example, it may include recommendation dimension information and recommendation frequency.

[0070] Recommendation dimension information: This information represents the source of the information display position corresponding to the recommendation request. Specifically, a recommendation system can be designed for multiple applications (apps). For example, an advertising recommendation system may recommend ads, but it could be used in multiple apps. Therefore, the advertising recommendation system can be integrated with these apps, and the source information would specify which app it originates from. Similarly, within an app, there may be different advertising recommendation scenarios. For instance, a news app might recommend ads in the entertainment section as well as the sports section, and the source information would specify which recommendation scenario it originates from. Furthermore, an app or a recommendation scenario can contain multiple different ad display positions, and the source information would again specify which ad display position it originates from.

[0071] Advertisers: Advertisers are the initiators of advertising campaigns, and are businesses that sell or promote their products and services online. Any business that promotes or sells its products or services can act as an advertiser. Advertisers publish advertising campaigns and pay the website owner according to the total number of marketing results and the unit price per result as stipulated in the advertising campaign.

[0072] Smart bidding advertising is a new bidding method where advertisers set optimization goals, such as mobile app downloads, activations, or payments, and the conversion cost they are willing to pay for those goals. The advertising platform automatically bids when an exposure opportunity exists, ensuring the actual conversion cost of the ad is close to the advertiser's expected conversion cost. Specifically, when an exposure opportunity exists, the advertising platform automatically bids for multiple ads and determines their ranking based on their bids, thus determining whether an ad receives the exposure. This process also affects whether the actual conversion cost after the ad receives exposure matches the advertiser's expected conversion cost. For example, in optimized cost per action (OCPA) advertising, the effective cost per mille (ECPM) is used as the metric for ad bidding ranking. By calculating the ECPM of each ad, ads with higher ECPMs are given priority for exposure.

[0073] The design concept of the embodiments of this application will be briefly introduced below.

[0074] In recommender systems, the accuracy of the recall and coarse-ranking stages largely determines the overall accuracy of the system. Each recommendation process involves recall, coarse-ranking, and fine-ranking. These stages may employ different sorting methods, leading to potential biases in the ranking of multimedia information. For example, a piece of multimedia information might have a high recommendation rate in the recall stage but a low rate in the fine-ranking stage. However, the recall and coarse-ranking stages determine which multimedia information advances to the fine-ranking stage. Research into the recommendation process in related technologies reveals that when recommending multimedia information to each account, the multimedia information that enters the candidate recommendation set within a short period is largely the same. Furthermore, the recommendation rate predicted by the fine-ranking model for multimedia information entering the fine-ranking stage shows high consistency. This means that multimedia information that was at the tail of the candidate recommendation set in the previous round is likely to remain at the tail in the next round of recommendations, meaning it won't be recommended to the user. However, the number of candidate recommendation sets is usually limited, and these multimedia information pieces occupy these slots for extended periods, leading to reduced utilization of the fine-ranking queue. Moreover, when the errors in the recall and coarse-ranking stages are large, other multimedia information that might be recommended to the user may not enter the candidate recommendation set, resulting in inaccurate recommendations and a relatively low accuracy rate for the recommendation system.

[0075] Based on this, this application provides a multimedia information recommendation method. In this method, a historical tail set of a target account is collected. Each historical tail set includes tail multimedia information from the candidate recommendation set corresponding to historical recommendation requests. When recommending multimedia information to the target account, an information filtering set is constructed based on at least one historical tail set and a set filtering strategy. After recall, multimedia information included in the information filtering set is filtered out from the recall set, and recommendations are made based on the filtered recall set. In this way, since the tail multimedia information in the candidate recommendation set usually has no chance of being recommended to the user, filtering the tail multimedia information avoids these tail multimedia information occupying candidate recommendation set slots for a long time, giving other multimedia information a chance to enter the candidate recommendation set, thus improving the utilization rate of the candidate recommendation set. At the same time, the generation of tail multimedia information may be caused by overestimation of the recommendation degree of these multimedia information during the recall or coarse-ranking stage. Filtering these essentially corrects the predictions made during the recall or coarse-ranking stage, thereby improving the accuracy of multimedia information recall and thus improving the overall accuracy of the recommendation system.

[0076] In this embodiment, tail set collection and filtering are performed for different recommendation dimensions, and different filtering strategy parameters can be adopted for different recommendation dimensions, making the method more applicable to different recommendation dimensions, thereby improving the accuracy of each recommendation dimension.

[0077] After introducing the design concept of the embodiments of this application, the following is a brief introduction to the application scenarios to which the technical solutions of the embodiments of this application can be applied. It should be noted that the application scenarios described below are only for illustrating the embodiments of this application and are not intended to limit the scope. In specific implementation, the technical solutions provided by the embodiments of this application can be flexibly applied according to actual needs.

[0078] The solution provided in this application can be applied to most multimedia information recommendation scenarios, such as advertising recommendation scenarios. Figure 1 The diagram shown is an application scenario diagram provided by an embodiment of this application. In this scenario, there are terminal devices 101, backend servers 102 and recommendation servers 103.

[0079] Terminal device 101 can be, for example, a mobile phone, tablet computer (PAD), laptop computer, desktop computer, smart TV, and smart wearable device. Terminal device 101 can have applications installed that can display multimedia information, such as browsers, video applications, news applications, or social applications. Users can view recommended multimedia information while browsing the application. The applications involved in this application embodiment can be software clients, web pages, mini-programs, etc., and the specific type of client is not limited.

[0080] Server 102 can be the backend server corresponding to the application installed on terminal device 101, and recommendation server 103 is a dedicated server used for multimedia information recommendation. Backend server 102 and recommendation server 103 can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms, but are not limited to these.

[0081] In one possible implementation, when a user browses on terminal device 101, a recommendation request for multimedia information can be triggered. For example, when a user opens a video page, terminal device 101 sends a video page request to backend server 102. If the video page contains an information display slot, then backend server 102 will send a recommendation request to recommendation server 103 to request the multimedia information to be displayed in that information display slot.

[0082] In another possible implementation, when a user browses on terminal device 101, a recommendation request for multimedia information recommendation can be sent directly to recommendation server 103, and a video page request for other content of the video page can be initiated to backend server 102.

[0083] The recommendation server 103 may include one or more processors 1031, a memory 1032, and an I / O interface 1033 for interacting with the terminal. Furthermore, the recommendation server 103 may be configured with a database 1034, which can be used to store the historical recommendation tails of each account and all multimedia information. The memory 1032 of the recommendation server 103 may also store program instructions for the multimedia information recommendation method provided in this application embodiment. When these program instructions are executed by the processor 1031, they can be used to implement the steps of the multimedia information recommendation method provided in this application embodiment to determine the multimedia information to recommend to the user, and then push the multimedia information to the target account.

[0084] In specific implementation, the backend server 102 and the recommendation server 103 can be separate and independent servers, or they can be different functional parts deployed on the same physical server. Alternatively, the backend server 102 and the recommendation server 103 can be the same server, meaning the multimedia information recommendation process can be implemented by the backend server applied on the terminal device 101. In this case, the terminal device 101 can directly initiate recommendation requests to the terminal device 101. This application embodiment does not limit the specific deployment method.

[0085] Terminal device 101, backend server 102, and recommendation server 103 can communicate directly or indirectly through one or more networks 104. The network 104 can be a wired network or a wireless network. For example, the wireless network can be a mobile cellular network or a Wireless-Fidelity (WIFI) network, or other possible networks. This application embodiment does not limit this.

[0086] It should be noted that in this embodiment of the application, the number of terminal devices 101 can be one or more, and similarly, the number of recommendation servers 103 can also be one or more. That is to say, there is no limitation on the number of terminal devices 101 or recommendation servers 103.

[0087] See Figure 2 The diagram shown is a schematic of the system architecture of the recommendation server 103. The recommendation server 103 includes an access layer module 103a, a fine ranking and filtering (Mixer) module 103b, a recall module 103c, a coarse ranking module 103d, and a data storage module 103e.

[0088] (1) Access layer module 103a

[0089] The access layer module 103a is responsible for the request access work of the recommendation server 103. It is used to obtain the user's recommendation request and perform certain protocol processing on the recommendation request in order to identify the content of the recommendation request and the protocol conversion.

[0090] (2) Mixer module 103b

[0091] The Mixer module 103b may further include a data acquisition submodule 103b1, an information filtering set construction submodule 103b2, a fine ranking submodule 103b3, and a data processing submodule 103b4. The data acquisition submodule 103b1 is used to obtain the historical tail set of the target account corresponding to the recommendation request from the data storage module 103e, and pass the historical tail set to the information filtering set construction submodule 103b2. The information filtering set construction submodule 103b2 constructs the information filtering set based on the historical tail set. The fine ranking submodule 103b3 is used to implement the fine ranking process in the recommendation process. The data processing submodule 103b4 is used to process the tail set of the current recommendation request and store it in the data storage module 103e.

[0092] (3) The recall module 103c and the coarse ranking module 103d are used to implement the recall and coarse ranking processes in the recommendation process, respectively.

[0093] (4) Data storage module 103e

[0094] The data storage module 103e is used to store various recommendation-related information, such as the historical tail set of each account. The data storage module 103e stores the tail set of requests initiated by each account in the most recent X hours after fine ranking, and stores the tail set of at most Y historical recommendation requests.

[0095] The methods and steps performed by the above modules will be described in detail in subsequent embodiments, so they will not be elaborated on here.

[0096] In one possible application scenario, this application embodiment recommends that various related data be stored using cloud storage technology. Cloud storage is a new concept that extends and develops from the concept of cloud computing. A distributed cloud storage system refers to a storage system that uses cluster applications, grid technology, and distributed storage file systems to aggregate a large number of storage devices (or storage nodes) of various types in the network through application software or application interfaces to work together and jointly provide data storage and business access functions to the outside world.

[0097] In one possible application scenario, to reduce communication latency, recommendation servers 103 can be deployed in various regions. Alternatively, for load balancing, different recommendation servers 103 can serve the regions corresponding to each terminal device 101. Multiple recommendation servers 103 can share data through blockchain, effectively forming a data-sharing system. For example, if terminal device 101 is located at location a, and its corresponding recommendation server 103 is the first recommendation server 103, then the first recommendation server 103 provides recommendation services to terminal device 101 at location a. If terminal device 101 is located at location b, and its corresponding recommendation server 103 is the second recommendation server 103, then the second recommendation server 103 provides recommendation services to terminal device 101 at location b.

[0098] Each recommendation server 103 in the data sharing system has a corresponding node identifier. Each recommendation server 103 can also store node identifiers of other recommendation servers 103 in the data sharing system, so that the generated blocks can be broadcast to other recommendation servers 103 in the data sharing system based on their node identifiers. Each recommendation server 103 can maintain a node identifier list as shown in the table below, storing the recommendation server 103 name and node identifier in this list. The node identifier can be an Internet Protocol (IP) address or any other information that can be used to identify the node; Table 1 only uses IP addresses as an example.

[0099] Server Name Node identifier Node 1 119.115.151.174 Node 2 118.116.189.145 … … Node N 119.124.789.258

[0100] Table 1

[0101] Of course, the methods provided in the embodiments of this application are not limited to... Figure 1 or Figure 2 The application scenarios shown can also be used in other possible scenarios, and this application embodiment does not impose any limitations. Figure 1 or Figure 2 The functions that each device in the application scenario shown can achieve will be described in subsequent method embodiments, and will not be elaborated on here.

[0102] Please see Figure 3 This is a flowchart illustrating the multimedia information recommendation method provided in this application embodiment. The method can... Figure 1 or Figure 2 The method is executed by either the recommended server 103 or the terminal device 101. Here, we will mainly take the recommended server 103 as an example for introduction. The process of this method is described as follows.

[0103] Step 301: Based on the target recommendation request of the target account, obtain at least one historical tail set of the target account; wherein, each historical tail set includes the last N multimedia information items in the candidate recommendation set corresponding to a historical recommendation request, sorted from high to low according to the recommendation degree, where N is a positive integer.

[0104] In this embodiment of the application, the process of step 301 can be implemented, for example, by the data acquisition submodule 103b1 included in the Mixer module 103b described above.

[0105] Specifically, when a user browses a page on a terminal device, a recommendation request will be triggered for that information display position because there may be information display positions on the page.

[0106] See Figure 4 As shown, when a user is playing a video, if the video playback page has an information display area, the terminal device will send a video playback request to the backend server to request data from the video playback page. This video playback request can carry a target recommendation request for the multimedia information to be displayed in the information display area, which the backend server can then send to the recommendation server. Alternatively, the terminal device can also send a target recommendation request to the recommendation server simultaneously with the video playback request to obtain the multimedia information to be displayed in that information display area.

[0107] certainly, Figure 4The information display space on the video playback page is just one possible scenario for displaying multimedia information. Information display spaces for displaying multimedia information can also exist in other scenarios or applications, such as news information pages in news applications, dynamic sharing pages in social platform applications, and music playback pages in music applications.

[0108] Considering that when recommending multimedia information to each account, the multimedia information that enters the candidate recommendation set in a short period of time is roughly the same, meaning that multimedia information that was at the end of the candidate recommendation set last time is likely to remain at the end in the next recommendation, essentially meaning that this multimedia information will not be recommended to the user. However, the number of candidate recommendation sets is usually limited, and these multimedia information pieces occupy the candidate recommendation set slots for a long time, preventing other multimedia information that might be recommended to the user from entering the candidate recommendation set. Therefore, after each fine-tuning, the multimedia information at the end of the candidate recommendation set can be stored, see [link to relevant documentation]. Figure 2 As shown, the tail set containing the tail multimedia information of each recommendation set is stored in the data storage module 103e. Each historical tail set includes the last N multimedia information in the candidate recommendation set corresponding to a historical recommendation request, sorted from high to low recommendation degree. The value of N can be set according to the specific situation, and this embodiment does not limit it.

[0109] In this embodiment of the application, when storing the tail set, it can be done according to preset storage parameters. The storage parameters can include the number Y of historical recommendation requests to be stored and the number of multimedia information in the tail set corresponding to each historical recommendation request, which is N mentioned above. The storage parameters can be set by the user or by the recommendation system based on recommendation attribute information.

[0110] See Figure 5 The diagram shown is a schematic of the storage structure of the tail set. Figure 5 Taking the storage of the tail set of a single account as an example, each account includes tail sets corresponding to Y historical recommendation requests, meaning there are Y tail sets, and each tail set includes the N multimedia information items that are ranked last during fine-tuning. Of course, the data storage module can also store all multimedia information in the fine-tuning queue corresponding to each historical recommendation request.

[0111] In one possible implementation, the recommendation attribute information can be the recommendation frequency to the target account. For example, when the recommendation frequency is high, the number of historical recommendation requests can be set to be large, and the number of multimedia information in each tail set can be set to be large.

[0112] In another possible implementation, the recommendation attribute information can be recommendation dimension information, and the tail set can be stored separately according to the different recommendation dimensions.

[0113] For example, when a recommendation system recommends multimedia information to different apps, the recommendation dimension can be each app. For an account, the tail set of historical recommendation requests corresponding to different apps can be stored separately, and the corresponding storage parameters Y and N can be set based on different apps. When storing, the values ​​of storage parameters Y and N corresponding to the recommendation dimension can be obtained, and the storage can be performed based on the values ​​of storage parameters Y and N.

[0114] For example, recommendation dimensions can also be set for different recommendation scenarios within the same app. For example, for a social app, information display positions can be set on the personal feed sharing page and the game feed sharing page respectively. The personal feed sharing page and the game feed sharing page belong to different recommendation scenarios and different recommendation dimensions. Therefore, corresponding storage parameters Y and N can be set based on different recommendation scenarios.

[0115] Of course, the recommendation dimension can also be specified to each information display position. That is, one recommendation dimension can be one information display position, and then the corresponding storage parameters Y and N can be set according to different information display positions.

[0116] In this embodiment of the application, taking the target account as an example, the target account can be any account in the recommendation system. When a recommendation request from the target account is received, the historical tail set of the target account can be obtained from the corresponding historical tail set based on the account identifier and recommendation attribute information of the target account.

[0117] For example, when the historical tail set of each account is stored separately by account, the historical tail set of the target account can be obtained based on the account identifier of the target account; when the historical tail set of each account is stored separately by different apps, the historical tail set corresponding to the app from which the target recommendation request originates can be obtained.

[0118] Specifically, since target recommendation requests may come from different apps, and different apps may use different protocols, upon receiving a target recommendation request, the protocol of the target recommendation request can be identified to determine the target site, i.e., the target app, that sent the target recommendation request. Then, based on the parsing method corresponding to the target site, the data of the target recommendation request can be parsed to obtain the source information of the information display position corresponding to the target recommendation request. Furthermore, based on the source information of the information display position, the recommendation dimension to which the target recommendation request belongs can be determined, and at least one historical tail set can be obtained from the historical tail set corresponding to that recommendation dimension.

[0119] Step 302: Based on at least one historical tail set and the set filtering strategy, construct an information filter set containing at least one multimedia information.

[0120] In this embodiment of the application, the process of step 302 can be executed, for example, by the information filtering set construction submodule 103b2 included in the Mixer module 103b described above.

[0121] Each recommendation process involves recall, coarse ranking, and fine ranking. These processes may employ different sorting methods, potentially leading to discrepancies in the ranking of multimedia information. For example, a piece of multimedia information might have a high recommendation rate during the recall phase but a low rate during the fine ranking phase. However, the recall and coarse ranking determine which multimedia information enters the fine ranking phase. Therefore, in this embodiment, after obtaining at least one historical tail set of the target account, an information filter set can be constructed for the current recommendation process based on a set filtering strategy. This information filter set represents the multimedia information that needs to be filtered after the recall phase during the current recommendation process.

[0122] Specifically, the filtering strategy indicates the selection method for multimedia information in the information filtering set. For example, the selection method can be to include multimedia information from all tail sets as the multimedia information included in the information filtering set; or, the selection method can be to sort the multimedia information according to a certain sorting method and then select the multimedia information that meets the conditions; or, the selection method can be to select a portion of multimedia information from each tail set to form the information filtering set.

[0123] In this embodiment of the application, the filtering strategy specifically includes multiple filtering parameters, which may include one or more of the following parameters:

[0124] (1) Threshold for the effective duration of the tail set

[0125] Considering that users' page browsing frequency is not fixed, there may be periods of high browsing frequency, during which the historical tail set is updated frequently. Therefore, the historical tail set stored in the data storage module will always be from the recent period. Conversely, during periods of low browsing frequency, the historical tail set may not be updated for a long time, potentially containing tail sets from much earlier periods. These tail sets are not very relevant to the current recommendation process and can therefore be ignored. Based on this, a valid duration threshold is set to help determine whether a tail set is valid. A historical tail set is considered valid when the time difference between the generation time of the historical recommendation request corresponding to it and the current time is not greater than the duration threshold; otherwise, it is invalid and will not be used when constructing the information filtering set.

[0126] (2) The number of historical tail sets Z

[0127] That is, when constructing the information filtering set, the historical tail set corresponding to the Z most recent historical recommendation requests is obtained as the construction basis.

[0128] (3) The amount of multimedia information M that needs to be acquired.

[0129] That is, when constructing the information filtering set, the M multimedia information items included in each tail set are selected as the construction basis.

[0130] Of course, the filtering strategy can be adjusted according to the specific situation. For example, the filtering strategy can also be to obtain the last M ads of each of the Z most recent historical recommendation requests, and form an information filtering set from these Z*M multimedia information.

[0131] In practice, due to inherent differences in different businesses and traffic volumes—for example, the recommendation frequency varies across different recommendation scenarios or apps—the optimal filtering parameters may differ for each traffic type. Consequently, different filtering strategies or parameters can be adopted for different recommendation attribute information. For instance, the filtering parameter values ​​in the filtering strategy can differ when the recommendation frequency to users varies, or the filtering parameter values ​​can also differ when the recommendation dimension information is different.

[0132] Specifically, the optimal filtering parameters for different recommended attribute information can be determined by setting multiple sets of different filtering parameters during the experimental phase and comparing the actual online performance of multiple sets of experiments. Of course, they can also be set based on empirical values.

[0133] In practice, the filtering strategy and parameters corresponding to the target recommendation request can be determined based on the recommendation attribute information. Then, an information filtering set is constructed based on the determined filtering strategy and parameters. Furthermore, considering that there may be duplicate multimedia information in the selected historical tail sets, the constructed information filtering set can be deduplicated. Finally, the deduplicated information filtering set is provided to the recall module.

[0134] Step 303: For the target recommendation request, obtain a recall set containing multiple multimedia information, and filter out the multimedia information included in the information filtering set from the recall set.

[0135] In this embodiment of the application, the process of step 303 can be achieved through... Figure 2 The recall module shown is used for this purpose. Specifically, when recalling multimedia information, different recall methods can be selected to recall multimedia information for the target recommendation request, and a recall set is formed based on the multiple recalled multimedia information.

[0136] The recall method can be any possible recall method, such as collaborative filtering based on user and multimedia information, or collaborative filtering based on the content displayed on the page where the current information display position is located and multimedia information.

[0137] During each recommendation process, recall and coarse ranking determine which multimedia information can enter the fine ranking stage. To prevent low-quality multimedia information from occupying the candidate recommendation set for fine ranking indefinitely, and to give other multimedia information that might receive higher recommendations in the fine ranking stage a chance to enter, after obtaining the recall set, multimedia information included in the information filtering set is removed from the recall set. This allows the remaining multimedia information to enter the fine ranking stage. For example, if 10,000 multimedia information items are recalled each time, and 1,000 are selected as candidate recommendations after coarse ranking, then after filtering the multimedia information in the information filtering set, these items, which would originally be among the 1,000 candidate recommendations, will be added to the candidate recommendation set after filtering. These new multimedia information items may receive better rankings in the fine ranking stage and thus be recommended to users.

[0138] Step 304: Based on the filtered recall set, obtain the candidate recommendation set corresponding to the target recommendation request, and determine the target multimedia information recommended to the target account from the candidate recommendation set.

[0139] In this embodiment of the application, after filtering the multimedia information in the information filtering set, a certain number of multimedia information can be selected from the filtered recall set through a coarse ranking process as the candidate recommendation set for this recommendation process. Then, based on the fine ranking process, the target multimedia information recommended for the target account is determined from the candidate recommendation set.

[0140] In this embodiment of the application, step 302 can be performed as follows: Figure 6 To achieve this, Figure 6 The flowchart for constructing the information filtering set is as follows.

[0141] Step 3021: Based on the set duration threshold, determine at least one set of historical tails in the target account that is in a valid state.

[0142] In this embodiment of the application, when determining whether a historical tail set is still valid, the time difference between the generation time of the historical recommendation request corresponding to the historical tail set and the current time can be obtained based on the generation time of the historical recommendation request, that is, the time when the user initiated the historical recommendation request. If the time difference is not greater than the duration threshold, the historical tail set is determined to be in a valid state; otherwise, if the time difference is greater than the duration threshold, the historical tail set is determined to be in an invalid state, and the historical tail set will not be used when constructing the information filtering set in the future.

[0143] Step 3022: Based on the recommendation attribute information of the target account or target recommendation request, obtain the values ​​of each filtering parameter in the filtering strategy that corresponds to the recommendation attribute information.

[0144] Specifically, when determining the values ​​of each filtering parameter, any of the following methods can be used:

[0145] In one possible implementation, if the corresponding filtering parameters are pre-set for each recommendation dimension, then the recommendation dimension to which the target recommendation request belongs can be determined first, and then the values ​​of each filtering parameter corresponding to the recommendation dimension in the filtering strategy can be obtained.

[0146] In another possible implementation, corresponding filtering parameters are pre-set for different recommendation frequencies. This allows us to first determine the first recommendation frequency for the target account, and then obtain the values ​​of each filtering parameter in the filtering strategy corresponding to the first recommendation frequency. The first recommendation frequency can refer to the overall recommendation frequency of the target account, which can be calculated from all historical recommendation records for that account in the recommendation system.

[0147] In another possible implementation, the values ​​of each filtering parameter in the filtering strategy corresponding to the second recommendation frequency can be obtained based on the second recommendation frequency of the target account in the recommendation dimension to which the target recommendation request belongs. In this implementation, the recommendation frequency of the account is counted separately for different recommendation dimensions, and then the corresponding filtering parameter information is obtained based on the determined recommendation frequency.

[0148] Step 3023: Based on the values ​​of the obtained filtering parameters, select at least one multimedia information from the historical tail set that is in a valid state to form an information filtering set.

[0149] Here, taking the filtering parameters including the number of historical recommendation requests Z and the number of multimedia information selected from each historical tail set as an example, we can select the historical tail sets corresponding to the last Z historical recommendation requests from the historical tail sets that are in a valid state, after sorting them in descending order of time difference. That is, we select the Z most recent historical tail sets from the historical tail sets in a valid state, and from each of these Z historical tail sets, we select the last M multimedia information after sorting them in descending order of recommendation degree. Then, based on the M multimedia information obtained from each of the Z historical tail sets, we construct an information filtering set.

[0150] See Figure 7 As shown, the data storage module stores the historical tail set corresponding to Y historical recommendation requests for the target account. Each historical tail set includes N multimedia information items. When the filtering parameter in the filtering strategy is to select the last M multimedia information items of the Z most recent historical recommendation requests, as follows... Figure 7 As shown, the Z most recent historical tail sets are selected from the historical tail sets that are in a valid state, and M multimedia information items are selected from each historical tail set, thus obtaining Z*M multimedia information items. After deduplication, the information filtering set is obtained.

[0151] In this embodiment of the application, the process of selecting target multimedia information in step 304 can be achieved as follows: Figure 8 To achieve this, Figure 8 This is a flowchart illustrating the fine sorting process, which can utilize methods such as... Figure 2 The fine arrangement submodule 103b3 shown is used to execute this process, which is described below.

[0152] Step 3041: Using the trained interaction rate acquisition model, based on the resource attribute information of each multimedia information in the candidate recommendation set, obtain the estimated interaction rate corresponding to each multimedia information.

[0153] The interaction rate acquisition model is trained using a large number of labeled training samples. Each training sample contains account information, resource attribute information of multimedia information, and information display position information, and is labeled with whether the account clicked on the multimedia information. In actual implementation, these training samples can be obtained from a historical database.

[0154] Among them, resource attribute information refers to attribute information related to multimedia information. For example, when the multimedia information is an advertisement, the resource attribute information may include information such as the advertisement name, the advertiser to which the advertisement belongs, the optimization goal of the advertisement, and the products involved in the advertisement.

[0155] Step 3042: Based on the obtained estimated interaction rates, determine the estimated amount of electronic resources that can be obtained after each multimedia information is displayed a set number of times on the information display position.

[0156] This example uses the OCPA (Optimal Cost of Pay) scheme in Smart Bidding Ads. The estimated electronic resource volume can be expressed as eCPM (Electronic Resource Payable). The formula for calculating eCPM is as follows:

[0157] ecpm=bid×pCTR×pCVR×lambda

[0158] Wherein, bid is the advertiser's expected conversion cost, pCTR is the predicted click-through rate obtained through the click-through rate prediction model, pCVR is the predicted conversion rate obtained through the conversion rate prediction model, and lambda is a real-time adjustment factor used to adjust the average conversion cost of the advertisement.

[0159] Of course, depending on the sorting method used, the estimated electronic resource quantity can also be other corresponding parameters, and this application embodiment does not limit this.

[0160] Step 3043: Sort the multiple multimedia information items according to the order of the obtained estimated electronic resource quantities from high to low.

[0161] Step 3044: Based on the ranking results, determine the target multimedia information from the candidate recommendation set.

[0162] Specifically, based on the ranking results, one or more multimedia information items with a higher estimated electronic resource volume can be selected from the candidate recommendation set as target multimedia information items and recommended to the target account.

[0163] In this embodiment of the application, by means of Figure 8 The process shown sorts the multimedia information in the candidate recommendation set. It then selects the tail set from the candidate recommendation set during this fine-tuning process and stores it in the data storage module for use in subsequent recommendation processes. This process can be achieved through... Figure 2The data processing submodule 103b4 in the Mixer module 103b shown is used for execution.

[0164] Specifically, based on the ranking results obtained from the fine ranking, the last N multimedia information items are selected from the candidate recommendation set to form the target tail set corresponding to the target recommendation request. The target tail set is then used as the historical tail set and added to the historical tail set library of the target account.

[0165] When storing historical tail sets, the target tail set corresponding to the target recommendation request can be stored in the corresponding target tail set based on the recommendation attribute information corresponding to the target recommendation request. For example, the historical tail set library can be divided according to the recommendation dimension information, thereby determining the recommendation dimension corresponding to the target recommendation request and adding the target tail set to the historical tail set library of the corresponding recommendation dimension.

[0166] In this embodiment of the application, when storing the historical tail set, relevant information for each tail set can also be stored. The relevant information may include one or more of the following information:

[0167] (1) Account identifier of the target account.

[0168] (2) The generation time of the target recommendation request can be used to help determine whether the historical tail set has expired.

[0169] (3) The information display position identifier corresponding to the target recommendation request can be used to help determine the recommendation dimension information of the target recommendation request in the dimension where the information display position is located.

[0170] (4) The information display scenario identifier corresponding to the target recommendation request can be used to help determine the recommendation dimension information of the target recommendation request in the dimension of the recommendation scenario.

[0171] (5) The target site identifier corresponding to the target recommendation request can be used to help determine the recommendation dimension information of the target recommendation request in the dimension of the site. The target site identifier indicates which site the target recommendation request comes from, that is, the APP.

[0172] Below, in conjunction with Figure 2 The architecture shown, taking advertising recommendation as an example, describes the technical solution process of the embodiments of this application. See also... Figure 9 The diagram shown illustrates the process of recommending advertisements.

[0173] Before recommending ads, relevant parameters need to be set in advance. These parameters can include storage parameters and filtering parameters for the filtering strategy. Storage parameters include the queue length for storing historical ad recommendation requests and the number of tail ads to be retrieved after each request's fine-sorting. Filtering parameters include the effective duration threshold, the maximum length of the queue for retrieving historical ad recommendation requests, and the number of tail ads to be retrieved.

[0174] S1: The user triggers an ad recommendation request to the access layer module of the recommendation server.

[0175] S2: The access layer module performs protocol conversion on the advertising recommendation request, transforming it into an internal recommendation request using a valid internal protocol, and then sends the internal recommendation request to the Mixer module.

[0176] S3: The data acquisition submodule of the Mixer module will retrieve the corresponding historical tail ad set from the data storage module based on the identity (ID) of the target account of the current request.

[0177] The data storage module stores the ad recommendation requests initiated by each account in the most recent X hours, the tail ad set after the fine-tuning stage, these ads are sorted by ecpm, and the information of at most Y ad recommendation requests is stored.

[0178] S4: The data acquisition submodule sends the historical tail advertisement set to the information filtering set construction submodule.

[0179] S5: The information filtering set construction submodule obtains the tail ad set, determines whether each tail ad set has expired according to the set time threshold, and then obtains the most recent Z requests and the tail M ads of each request according to the specified filtering parameters. Based on these Z*M ads, the ad filtering set is formed after deduplication, and the constructed ad filtering set is passed to the ad recall module.

[0180] S6: After the ad recall process is completed, the recall set is traversed once, and ads that are in the recall set and in the ad filtering set are filtered out in the recall set.

[0181] S7: After coarse sorting, return the candidate ad set to the Mixer module.

[0182] S8: After the fine-ranking process is completed, send the information of the tail advertisements of this fine-ranking process to the data processing module.

[0183] The ad collection information may include account ID, timestamp of ad recommendation request, ad slot ID, scene ID, site ID, and ad ID of each ad sorted by eCPM.

[0184] In the fine-ranking stage, each advertisement is sorted according to a set sorting method, such as sorting based on ecpm. The advertisement with the highest ranking is selected and returned to the access layer, which then sends it to the user side to display the selected advertisement to the user.

[0185] S9: The data storage module updates the tail ad collection information to the data storage module and determines whether the previously stored tail ad collection has expired. If any data has expired, the corresponding data is deleted to ensure that the data in the current storage queue is within the validity period and has a maximum of Y requests.

[0186] For advertising recommendation systems, filtering historical tail ads gives other ads the opportunity to enter the fine-ranking queue. These ads may not have achieved good scores in the recall or coarse-ranking stages, but they may have achieved good scores in the fine-ranking stages and may be recommended to users, thereby improving the recommendation accuracy of the advertising platform and thus increasing the platform's Gross Merchandise Volume (GMV).

[0187] Please see Figure 10 Based on the same inventive concept, this application also provides a multimedia information recommendation device 100, which includes:

[0188] The acquisition unit 1001 is used to acquire at least one historical tail set of the target account based on the target recommendation request of the target account; wherein, each historical tail set includes the last N multimedia information in the candidate recommendation set corresponding to the historical recommendation request, sorted from high to low according to the recommendation degree, where N is a positive integer;

[0189] Construction unit 1002 is used to construct an information filter set containing at least one multimedia information based on at least one historical tail set and a set filtering strategy;

[0190] The filtering unit 1003 is used to obtain a recall set containing multiple multimedia information for a target recommendation request, and filter out the multimedia information included in the information filtering set from the recall set.

[0191] The recommendation unit 1004 is used to obtain a candidate recommendation set corresponding to the target recommendation request based on the filtered recall set, and to determine the target multimedia information recommended to the target account from the candidate recommendation set.

[0192] Optionally, construction unit 1002 is specifically used for:

[0193] Based on a set duration threshold, at least one set of historical tails is identified as being in a valid state; wherein, a set of historical tails is in a valid state when the time difference between the generation time of the historical recommendation request corresponding to a set of historical tails and the current time is not greater than the duration threshold.

[0194] Based on the recommendation attribute information of the target account or target recommendation request, obtain the values ​​of each filtering parameter in the filtering strategy that corresponds to the recommendation attribute information;

[0195] Based on the values ​​of the obtained filtering parameters, at least one multimedia information is selected from the set of historical tails that are in a valid state to form an information filtering set.

[0196] Optionally, construction unit 1002 is specifically used for:

[0197] Based on the recommendation dimension information to which the target recommendation request belongs, obtain the values ​​of each filtering parameter in the filtering strategy that corresponds to the recommendation dimension information; whereby the recommendation dimension information is used to characterize the source information of the information display position corresponding to the target recommendation request;

[0198] Based on the first recommendation frequency of the target account, obtain the values ​​of each filtering parameter in the filtering strategy corresponding to the first recommendation frequency;

[0199] Based on the second recommendation frequency of the target account in the recommendation dimension to which the target recommendation request belongs, obtain the values ​​of each filtering parameter in the filtering strategy corresponding to the second recommendation frequency.

[0200] Optionally, each filtering parameter includes the number Z of historical recommendation requests corresponding to the historical tail set, and the number M of multimedia information selected from each historical tail set, where M is a positive integer and M≤N; then the construction unit 1002 is specifically used for:

[0201] From the set of historical tails that are in a valid state, select the last Z sets of historical recommendation requests after sorting them in descending order of time difference; where the time difference is the time difference between the generation time of the historical recommendation request corresponding to each set of historical tails and the current time.

[0202] From the last Z sets of historical tails, select the last M multimedia information items sorted from highest to lowest recommendation level;

[0203] Based on the M multimedia information obtained from each of the Z historical tail sets, an information filtering set is constructed.

[0204] Optionally, the acquisition unit 1001 is specifically used for:

[0205] The target recommendation request is identified by protocol to determine the target site that sent the request.

[0206] Based on the parsing method corresponding to the target site, the target recommendation request is parsed to obtain the source information of the information display position corresponding to the target recommendation request;

[0207] Based on the source information of the information display position, determine the recommendation dimension to which the target recommendation request belongs, and obtain at least one historical tail set from the historical tail set corresponding to the recommendation dimension.

[0208] Optional, recommended unit 1004, specifically used for:

[0209] The trained interaction rate acquisition model is used to obtain the estimated interaction rate of each multimedia information based on the resource attribute information of each multimedia information in the candidate recommendation set.

[0210] Based on the obtained estimated interaction rates, the estimated amount of electronic resources that can be obtained after each multimedia message is displayed a set number of times in the information display position is determined.

[0211] Based on the obtained estimated electronic resource quantities, the multiple multimedia information items are sorted from high to low.

[0212] Based on the ranking results, the target multimedia information is determined from the candidate recommendation set.

[0213] Optionally, the device further includes a collection update unit 1005, used for:

[0214] Based on the ranking results, the last N multimedia information items in the candidate recommendation set are selected to form the target tail set corresponding to the target recommendation request;

[0215] Add the target tail set as a historical tail set to the historical tail set library of the target account.

[0216] Optionally, set update unit 1005 is also used for:

[0217] Add the relevant information of the target tail set to the historical tail set database of the target account; wherein, the relevant information includes one or more of the following combinations:

[0218] The target account's account identifier;

[0219] The time when the target recommendation request is generated;

[0220] The identifier for the information display position corresponding to the target recommendation request;

[0221] The information display scenario identifier corresponding to the target recommendation request;

[0222] The target site identifier corresponding to the target recommendation request.

[0223] This device can be used to perform Figures 3-9 The method shown in the illustrated embodiment is relevant here; therefore, the functions that each functional module of the device can achieve can be referred to. Figures 3-9 The embodiments shown are described in detail below.

[0224] Please see Figure 11 Based on the same technical concept, this application also provides a computer device 110, which may include a memory 1101 and a processor 1102.

[0225] The memory 1101 is used to store computer programs executed by the processor 1102. The memory 1101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function, etc.; the data storage area may store data created based on the use of the computer device, etc. The processor 1102 may be a central processing unit (CPU), or a digital processing unit, etc. This application embodiment does not limit the specific connection medium between the memory 1101 and the processor 1102. This application embodiment... Figure 11 The memory 1101 and the processor 1102 are connected via a bus 1103, and the bus 1103 is in Figure 11 The connections between other components are shown in bold and are for illustrative purposes only, not as limiting information. The bus 1103 can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0226] Memory 1101 may be volatile memory, such as random-access memory (RAM); memory 1101 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1101 may be any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1101 may be a combination of the above-described memories.

[0227] Processor 1102 is configured to execute, when calling a computer program stored in memory 1101, such as Figures 3-9 The method performed by the device in the illustrated embodiment.

[0228] In some possible implementations, various aspects of the methods provided in this application can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the methods according to the various exemplary embodiments of this application described above. For example, the computer device may perform actions such as... Figures 3-9 The method performed by the device in the illustrated embodiment.

[0229] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0230] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0231] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A multimedia information recommendation method characterized by comprising: The method comprises: obtaining at least one historical tail set of the target account based on a target recommendation request of the target account; wherein each of the historical tail sets comprises the last N multimedia information in a candidate recommendation set corresponding to a historical recommendation request in a descending order of recommendation degree, N being a positive integer; determining a historical tail set in an effective state in the at least one historical tail set based on a set time threshold; wherein when a time difference between a generation time of a historical recommendation request corresponding to a historical tail set and a current time is not greater than the time threshold, the historical tail set is in the effective state; obtaining the value of each filtering parameter corresponding to the recommendation attribute information in the set filtering strategy based on the recommendation attribute information of the target account or the target recommendation request; selecting at least one multimedia information from the historical tail set in the effective state based on the obtained value of each filtering parameter to form an information filtering set; obtaining a recall set containing a plurality of multimedia information for the target recommendation request, and excluding the multimedia information included in the information filtering set from the recall set; obtaining a candidate recommendation set corresponding to the target recommendation request based on the filtered recall set, and determining the target multimedia information recommended for the target account from the candidate recommendation set.

2. The method of claim 1, wherein, obtaining the value of each filtering parameter corresponding to the recommendation attribute information in the filtering strategy based on the recommendation attribute information of the target account or the target recommendation request comprises any one of the following manners: obtaining the value of each filtering parameter corresponding to the recommendation dimension information in the filtering strategy according to the recommendation dimension information to which the target recommendation request belongs; wherein the recommendation dimension information is used to represent the source information of the information display position corresponding to the target recommendation request; obtaining the value of each filtering parameter corresponding to the first recommendation frequency in the filtering strategy according to the first recommendation frequency of the target account; obtaining the value of each filtering parameter corresponding to the second recommendation frequency in the filtering strategy according to the second recommendation frequency of the target account in the recommendation dimension to which the target recommendation request belongs.

3. The method of claim 1, wherein, The plurality of filtering parameters comprises the number Z of historical recommendation requests corresponding to the historical tail set, and the number M of multimedia information selected from each historical tail set, wherein M is a positive integer and M≤N; Then, the selecting of the at least one multimedia information from the historical tail set in the effective state based on the obtained value of each filtering parameter to form the information filtering set comprises: selecting the last Z historical tail sets corresponding to the last Z historical recommendation requests in descending order of time difference from the historical tail set in the effective state; wherein the time difference is the time difference between the generation time of the historical recommendation request corresponding to each historical tail set and the current time; respectively selecting the last M multimedia information in each of the last Z historical tail sets in descending order of recommendation degree; Construct the information filtering set based on M multimedia information respectively obtained from each of the Z historical tail sets.

4. The method according to any one of claims 1 to 3, characterized in that The target recommendation request based on the target account is used to obtain at least one historical tail set of the target account, including: The target recommendation request is protocol-identified to determine a target site sending the target recommendation request; Based on the analysis method corresponding to the target site, the data of the target recommendation request is analyzed to obtain the source information of the information display position corresponding to the target recommendation request; Based on the source information of the information display position, the recommendation dimension to which the target recommendation request belongs is determined, and the at least one historical tail set is obtained from the historical tail set corresponding to the recommendation dimension.

5. The method according to any one of claims 1 to 3, wherein From the candidate recommendation set, the target multimedia information recommended for the target account is determined, including: An interaction rate acquisition model is trained, and based on the resource attribute information of each multimedia information in the candidate recommendation set, the pre-estimated interaction rate corresponding to each multimedia information is obtained respectively; Based on the obtained pre-estimated interaction rate, the pre-estimated electronic resource amount that each multimedia information can obtain after a set number of displays on the information display position is determined respectively; Based on the order from high to low of the obtained pre-estimated electronic resource amount, the multiple multimedia information is sorted; According to the sorting result, the target multimedia information is determined from the candidate recommendation set.

6. The method of claim 5, wherein, After the multiple multimedia information is sorted according to the obtained electronic resource amount, the method further includes: Based on the sorting result, the last N multimedia information in the sorting is selected from the candidate recommendation set to constitute the target tail set corresponding to the target recommendation request; The target tail set is added to the historical tail set library of the target account as a historical tail set.

7. The method of claim 6, wherein, The method further includes: The related information of the target tail set is added to the historical tail set library of the target account; wherein the related information includes one or a combination of the following information: The account identifier of the target account; The generation time of the target recommendation request; The information display position identifier corresponding to the target recommendation request; The information display scene identifier corresponding to the target recommendation request; The target site identifier corresponding to the target recommendation request.

8. A multimedia information recommendation apparatus characterized by comprising: The device includes: An acquisition unit is configured to obtain at least one historical tail set of a target account based on a target recommendation request of the target account; wherein each historical tail set includes the last N multimedia information in a candidate recommendation set corresponding to a historical recommendation request when sorted from high to low according to the recommendation degree, and N is a positive integer; The construction unit is configured to determine a history tail set in an active state from the at least one history tail set based on a set time length threshold; when a time difference between a generation time of a history recommendation request corresponding to the history tail set and a current time is not greater than the time length threshold, the history tail set is in the active state; obtain a value of each filter parameter corresponding to recommendation attribute information of the target account or the target recommendation request in a set filter strategy based on the recommendation attribute information; and select at least one multimedia information from the history tail set in the active state based on the obtained value of each filter parameter to form an information filtering set; The screening unit is configured to obtain a recall set containing a plurality of multimedia information for the target recommendation request, and screen out the multimedia information included in the information filtering set from the recall set; The recommendation unit is configured to obtain a candidate recommendation set corresponding to the target recommendation request based on the screened recall set, and determine target multimedia information recommended for the target account from the candidate recommendation set. 9.A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7. 10.A computer storage medium having computer program instructions stored thereon, wherein: The computer program instructions are executed by a processor to implement the steps of the method in any one of claims 1 to 7.

Images